I didn't know this technique existed at all, but it's really cool. Anyone have a good explanation?

Here is one that I found: https://scaniverse.com/news/intro-gaussian-splats

Seems to me like it's a way to capture a bunch of 2D pictures of a scene, but instead of turning them into meshes, you position them in a 3D point cloud but then use algorithms to calculate their position, angle, field of view, etc. and them seamlessly blend them together with other nearby pictures to create the illusion of depth. Is that right?

The ability to overlay and merge multiple .ply files [1] is a great feature if you can match reference points and scale from the original scan then it can fill in detail gaps by running a few capture sessions. GSplats still need a lot of 'pixel' level cleanup to get the sample results quality level still beats point clouds any day of the week!

[1] https://github.com/playcanvas/supersplat/wiki/Editing-Scenes

This is amazing and of course reminds me of Sketchfab. Nice work. I feel like people don't really get Gaussian Splats and what the advantages are over polygons.

I wish it worked better on Firefox mobile, but the examples look really good.

the bike in the demo

https://superspl.at/view?id=97c2720d

What software do you use to generate them?